A relaxation algorithm for speaker verification is presented in this paper. This verification system uses a nonlinear probabilistic framework to paralelly and iteratively update the labelings of the input speech feature vectors until no further improvements are found. The implementation is tested on a commercial speech corpus TI46. The equal error rates obtained from the proposed approach are significantly more favorable than those from the conventional VQ-based method.
|Number of pages||4|
|Publication status||Published - 2020|
|Event||Workshop on Speaker Recognition and its Commercial and Forensic Applications, RLA2C 1998 - Avignon, France|
Duration: 20 Apr 1998 → 23 Apr 1998
|Conference||Workshop on Speaker Recognition and its Commercial and Forensic Applications, RLA2C 1998|
|Period||20/04/98 → 23/04/98|